Académique Documents
Professionnel Documents
Culture Documents
C
2011 by the Society for Marine Mammalogy
DOI: 10.1111/j.1748-7692.2010.00456.x
Assessing population parameters and trends
of Guiana dolphins (Sotalia guianensis): An eight-year
mark-recapture study
MAURICIO CANTOR
Laborat orio de Mamferos Aqu aticos,
Departamento de Ecologia e Zoologia,
Centro Ci encias Biol ogicas,
Universidade Federal de Santa Catarina,
P. O. Box 5102, CEP 88040-970,
Florian opolis, Santa Catarina, Brazil
and
Instituto Baleia Jubarte,
Rua Bar ao do Rio Branco, 26,
CEP 45900-000, Caravelas, Bahia, Brazil
E-mail: m.cantor@ymail.com
LEONARDO L. WEDEKIN
Instituto Baleia Jubarte,
Rua Bar ao do Rio Branco, 26,
CEP 45900-000, Caravelas, Bahia, Brazil
F ABIO G. DAURA-JORGE
Laborat orio de Mamferos Aqu aticos,
Departamento de Ecologia e Zoologia,
Centro Ci encias Biol ogicas,
Universidade Federal de Santa Catarina,
P. O. Box 5102, CEP 88040-970,
Florian opolis, Santa Catarina, Brazil
MARCOS R. ROSSI-SANTOS
Instituto Baleia Jubarte,
Rua Bar ao do Rio Branco, 26,
CEP 45900-000, Caravelas, Bahia, Brazil
PAULO C. SIM OES-LOPES
Laborat orio de Mamferos Aqu aticos,
Departamento de Ecologia e Zoologia,
Centro Ci encias Biol ogicas,
Universidade Federal de Santa Catarina,
P. O. Box 5102, CEP 88040-970,
Florian opolis, Santa Catarina, Brazil
63
64 MARINE MAMMAL SCIENCE, VOL. 28, NO. 1, 2012
ABSTRACT
This study represents the rst attempt to study the population dynamics of
Guiana dolphins (Sotalia guianensis), by evaluating a set of demographic parame-
ters. The population of the Caravelas River estuary, eastern Brazil, was systemat-
ically monitored through a long-term mark-recapture experiment (20022009).
Abundance estimates revealed a small population (57124 dolphins), comprised
of resident dolphins and individuals that temporarily leave or pass through the
study area. Temporary emigration from the estuary to adjacencies (
= 0.33
0.07 SE) and return rate (1
N to 27
30
S, 39
30
W) and
adjacent coastal areas, spanning more than 700 km
2
. It is located on the Abrolhos
Bank, an extension of the continental shelf on the eastern coast of Brazil (Fig. 1). The
region is highly heterogeneous, characterized by a mosaic of open waters protected
by coral reef barriers, mangrove forests with channels, sandy beaches, and banks of
shallow waters.
Data Collection and Sampling Design
Amark-recapture experiment was conducted using individual recognition through
photo identication. For each animal a capture history was created, where a capture
event in a sampling occasion was denoted as 1 (here a photographic record) and a
not capture as 0. From maximum likelihood estimation procedures, we derived the
population parameters that maximize the likelihood of observed our capture histories
frequencies (Lebreton et al. 1992).
66 MARINE MAMMAL SCIENCE, VOL. 28, NO. 1, 2012
Figure 1. Caravelas River Estuary, coastal adjacent waters and coral reefs in the Abrolhos
Bank, eastern coast of Brazil. Black dots indicate groups with identied Guiana dolphins.
Data collection was carried out during surveys using a 5 m inatable boat
(50 hp outboard engine) between 2002 and 2005, and using a 12 m wooden vessel
(33 hp inboard engine) from 2006 to 2009. Routes aimed to cover the study area ho-
mogeneously: each covered an average of 30 nmi (56.6 km), including departing and
arriving at the estuary, and aimed to sequentially sample four different subregions
within the study area (north, east, southeast, and south; Fig. 1). Given the period
required to cover the whole area, four consecutive sampling days were pooled and
treated as a capture occasion.
Groups of dolphins were searched at slow speed (up to 5 kn) by two to ve
observers, alternating in three positions to cover 180 degrees of the visual eld. For all
sightings, we recorded geographic coordinates, time, and the number of individuals.
We attempted to photograph the dorsal n of all dolphins in the group, taking as
many photos as possible of both sides and without individual preferences. Because
individuals cannot be recognized during the data collection, the capture effort was
assumed equal for all animals. From April 2002 to July 2004, photographs were
taken using a SLR camera (Nikon N-90), equipped with 300, 70300, and 120
400 mm lenses. In August 2004 we switched to digital SLR cameras (Nikon D70
and more recently D80) and the same lenses.
Photo-Identication
Guiana dolphins were identied by natural markings on the dorsal n, a technique
widely used for individual recognition of cetacean species (Hammond et al. 1990),
including Guiana dolphins (e.g., Flores 1999). During 2002 and 2004, the slide
lms were projected onto a white surface to produce prole drawings of each marked
CANTOR ET AL.: POPULATION DYNAMICS OF GUIANA DOLPHINS 67
dorsal n (adapted from Defran et al. 1990; see Rossi-Santos et al. 2007). The
digital catalog with photos from the entire period was periodically reviewed and
included only high quality photographs (good focus and perpendicular angle). To
minimize misidentication, the photo analysis was restricted to individuals with
a very distinctive dorsal n (at least one long-lasting notch on the trailing edge,
Hammond et al. 1990). Calves and nonidentiable individuals (intact or indistinctive
dorsal n, or captured only on low-quality photographs) were not included in our
analyses.
Discovery Curves
To visualize differences in sampling effort among years and to provide a valid com-
parison of the number of identied individuals under varying sampling effort, dis-
covery curves were generated using the same method as for sample-based rarefaction
curves (Gotelli and Colwell 2001). These are more robust surrogates of the traditional
cumulative curves, representing the means of repeated resampling of the capture
occasions at random and plotting the average number of marked individuals (Gotelli
and Colwell 2001). Resampling was done by Monte Carlo methods, setting 1,000
iterations and the expected curves were developed as a function of number of samples
(Mao Tau) in the software EstimateS (Colwell RK, http://purl.oclc.org/estimates).
Mark-Recapture Analysis
Mark-recapture analyses were conducted using Program MARK (White and
Burnham 1999). We used CormackJollySeber (CJS) models for open popula-
tions (Cormack 1964, Jolly 1965, Seber 1965) to estimate apparent survival () and
capture probability (p). Departing from the full time varying CJS model, a set of can-
didate models were developed to test different effects on the estimated parameters:
no variation (), time dependence (t); time since marking, i.e., age-dependence (a2);
and the set of dolphins identied in a given occasion (cohort) (Lebreton et al. 1992).
Because sampling effort varied during the study, we also introduced a period effect in
the models, representing the variation of effort among the rst 3 yr and the last 4 yr.
The rst 3 yr were characterized by broader sampling efforts with analog cameras,
whereas in the later years, effort was reduced and digital cameras were used.
Mark-recapture models make the following assumptions (see Amstrup et al. 2005):
(1) marks are not lost during the study, (2) marks are correctly recognized on
recapture, (3) individuals are instantly released after being marked, (4) intervals
between sampling occasions are longer than the duration of the sample, (5) all
individuals observed during a given sampling occasion have the same probability of
surviving to the next one, (6) study area does not vary, and (7) marked and unmarked
animals have equal capturability. We relied on the validation of open population
assumptions for dolphins discussed by Silva et al. (2009). In addition, we validated
the assumptions of equal probabilities of capture and survival by the goodness-
of-t test using program RELEASE (Lebreton et al. 1992). Even if one or more
assumptions failed, the occurrence and magnitude of the resulting extra-binomial
variation (overdispersion, see Anderson et al. 1994) was evaluated and accounted for.
We estimated the variance ination factor ( c ) through the Median c and bootstrap
approaches, available in Program MARK (White and Burnham 1999), and used the
highest value to adjust the lack of t of the models.
68 MARINE MAMMAL SCIENCE, VOL. 28, NO. 1, 2012
Pollocks Robust Design (RD) (Pollock 1982; Kendall et al. 1995, 1997) was
applied to assess population size and emigration pattern. A year was considered as
the primary period and used to estimate apparent survival. The 4 d pooled capture
occasions within each year were set as the secondary periods and used to estimate
abundance through closed-population models. The modications on RD by Kendall
et al. (1997), which allow for an animal in the population to be unavailable for
capture at a given time, were used to estimate temporary emigration.
The models based on the RD were composed of the following parameters:
t
=
the apparent survival probability from primary period t to (t + 1); p
st
, c
st
= the
probability that an individual available for capture in period t would be recaptured
in the secondary sample s of the primary period t;
t
= the probability that an
individual would be unavailable for capture during primary period t given that it
was unavailable or available (respectively) in the previous period (i.e., the probability
of temporary emigration). Population size (
N) was estimated using the full parame-
terization of maximum likelihood available in MARK (White and Burnham 1999).
The following models were designed using the notation provided by Kendall et al.
(1997): Markovian emigration models (
= 0).
The eight classical closed population models (Otis et al. 1978) used in the RD
(Kendall et al. 1995) have other specic assumptions for the secondary period (for
assessment of assumptions to cetacean data, see Wilson et al. 1999, Bearzi et al. 2008):
(1) demographic closure; (2) lack of behavioral responses to capture procedure, i.e.,
animals do not respond to being captured in a way that affects their subsequent
probability of recapture; and (3) homogeneity of capture probabilities, i.e., within
a sampling occasion, all animals in the population have equal probability of being
captured. Behavioral response models were discarded because we assumed a priori
there were no reactions to the capture procedure involving photo-identication (trap
dependence). Thus, the capture probability (p) was considered equal to the recapture
probability (c). Individual heterogeneity was included using the Pledgers mixture
models, with two mixtures of capture probabilities (2 pi) (Pledger 2000). However,
heterogeneity in capture probabilities was included only in those models with no
emigration. The inuence of time was tested for all parameters, both in primary
periods (t) and secondary periods (s).
For the CJS approach, the most appropriate model was selected through the Quasi-
Akaike Information Criterion (QAIC
c
, Anderson et al. 1994), while the Akaike
Information Criterion (AIC
c
) was used for RD models. In all cases, specic biological
hypotheses between nested models were tested using likelihood ratio tests (LRT,
Burnham and Anderson 2002). Normalized QAIC
c
weightsor AIC
c
weights for
RD modelswere used to measure the support for a given model relative to the
others. Parameter estimates were averaged across all models based on QAIC
c
or
AIC
c
weights (Burnham and Anderson 2002).
Trends in Abundance
Because the abundance estimates (
N) refer exclusively to the well-marked animals
in the population, we corrected this to include the unmarked individuals in the total
abundance estimates (
N
,
and the variances of total population size were estimated using the delta method
(Seber 1982) as:
var
_
N
_
=
N
2
_
var
_
N
_
N
2
+
1
n
_
,
where n is the total number of individuals from which
N was estimated. Log-normal
condence intervals for total population size (see Burnhamet al. 1987) were calculated
as:
C = exp
_
z
2
_
ln
_
1 +
_
CV
_
N
__
2
_
_
,
where z is the normal deviate, = 0.05, and CV is the coefcient of variation.
To determine the probability of detecting a linear population trend in the corrected
abundance estimates, we performed a statistical power analysis (Fairweather 1991).
By denition, a trend exists when the regression has a slope signicantly different
from zero (see Gerrodette 1987). Power analysis provides the probability of correctly
rejecting the null hypothesis of constant population when it is actually increasing or
decreasing (1 , where is the probability of Type 2 error).
70 MARINE MAMMAL SCIENCE, VOL. 28, NO. 1, 2012
To explore the monitoring efciency, we simulated two kinds of changes in the
population: a precipitous decline of 50% in the abundance during the entire mon-
itoring period (e.g., Taylor et al. 2007) and a modest decline of 5% per year (e.g.,
Lukoschek and Chilvers 2008). We analyzed the effort necessary to detect these
changes with an acceptable power of 80% certainty (see Taylor et al. 2007).
Analyses were conducted using the software TRENDS (Gerrodette 1987, 1993),
setting the parameters as follows: the signicance criterion =0.05; one-tailed test
(because we are only concerned with a decrease); equal intervals between sampling
occasions and exponential type of change. Corrected abundance estimates (
N
) were
transformed to the natural logarithm before performing the regression. Variance
usually uctuates according to some function of the abundance (Gerrodette 1993),
and for mark-recapture estimates, the CVon abundance is expected to be proportional
to the square root of abundance (Gerrodette 1987). However, it can also vary due
to the species and the method used. To test this, we plotted CVs of each year
against
_
N
and (
_
N
)
1
(cf . Gerrodette 1987). Since no clear relationship arose,
we assumed CVs were constant with respect to abundance and thus calculated the
overall CVof the monitoring period by averaging the annual CVs (Gerrodette 1993).
RESULTS
Sampling Effort
From 2002 to 2009, sampling was performed on 389 d and covered more than
13,425 nmi. We recorded 380 groups, totaling 1,702 sighted dolphins (10% calves)
(Table 1). A total of 108 individuals were identied with good-quality photographs,
and 12 were resighted in all years. Greater sampling effort was employed between
2002 and 2004 than in the later years. The number of individuals with long-lasting
marks included in the analysis varied among years (Table 1). The rarefaction curves
for 2002 to 2004 tended to stabilize with narrower condence intervals (Fig. 2A,
B). The opposite was found for surveys from 2005 on, when all condence intervals
were wider, and the curves were nonasymptotic (Fig. 2BD). The number of new
photo identied individuals per year uctuated from 15 (2004) to 40 (2007), while
the recapture rate varied from 55% (2007) to 86% (2004). From 30% to 64%
of photo-identied dolphins in one year were recaptured in the subsequent year
(Table 2).
Model Selection for Estimating Population Parameters
From the results of all components of the goodness-of-t test, we did not violate
the assumptions of equal probabilities of capture (TEST 2; P = 0.778) and survival
(TEST3; P=0.522). The most parameterized CJS model tted our data satisfactorily.
The bootstrap approach estimated the highest variation ination factor, which was
marginally above 1 and indicated no substantial overdispersion ( c = 1.25). It may
be caused by the presence of temporary emigrants in the population, or other source
of heterogeneity of capture probability (see further). Despite the lack of evidence for
such an effect, we adjusted the models with c to correct any potential violation of
assumptions or intrinsic variations in the population.
In general, the CJS models with time (Table 3: models 15, 13, 12) or age-
dependence (models 6, 3) of survival and models that allowed this parameter to vary
with sampling effort (models 14, 9, 5) poorly t the data or were not parsimonious.
CANTOR ET AL.: POPULATION DYNAMICS OF GUIANA DOLPHINS 71
s
n
i
h
p
l
o
D
d
e
k
r
a
M
f
o
r
e
b
m
u
N
Capture Occasions
A
C D
B
55
45
35
25
15
5
55
45
35
25
15
5
1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 11 13 15 17 19
2007
2006
2008
2009
2004
2005
2002
2003
Figure 2. Sample-based rarefaction curves of cumulative photo-identied Guiana dolphins
in the Caravelas River estuary from 2002 to 2009. Black curves represent the Mao Tau
estimates and gray curves are the respective 95% condence intervals.
Similarly, models with capture probability dependent on time (models 12, 5, 4, 3),
period (models 15, 10) or both (models 16, 14, 11, 7) also provided a poor t to the
data. Thus, the best CJS model had constant survival and capture probability varying
in relation to cohort (model 1). The LRT agreed with the best model selected by
QAIC
c
, when comparing nested models. Moreover, there were no additional effects
of period (model 2; P =0.282) or time (model 11; P =0.333) on capture probability.
RD models not accounting for temporary emigration poorly t the data (Table 4:
models 912), and emigration probabilities seemed to follow a Random Movement
Table 2. M-array of capture-recapture data used for open-population models. R(i) =number
of individuals marked (photo-identied for the rst time) for each occasion i; m(i, j) =number
of individuals marked in occasion i and recapture for the rst time in a capture occasion j;
r(i) = total of individuals marked in occasion i and recaptured throughout the subsequent
occasions.
m(i, j)
Occasions R(i) 2003 2004 2005 2006 2007 2008 2009 r(i)
2002 37 24 1 0 1 3 1 0 30
2003 31 10 10 4 2 0 0 26
2004 15 9 3 1 0 0 13
2005 34 10 7 1 2 20
2006 24 12 4 2 18
2007 40 12 10 22
2008 26 15 15
72 MARINE MAMMAL SCIENCE, VOL. 28, NO. 1, 2012
Table 3. CormackJollySeber models for survival () and capture probability (p) ranked
by the lowest QAIC
c
. QAIC
c
weight indicates the strength of evidence for a given model.
Model notation: constant parameter (), time (t), age (a2), cohort or period dependence.
QAICc No.
Model QAICc QAICc weights Likelihood parameters
1 {() p(cohort)} 411.621 0 0.468 1 8
2 { () p(cohort period)} 413.487 1.87 0.184 0.393 10
3 { (a2) p(t)} 413.487 1.87 0.184 0.393 9
4 { () p(t)} 415.639 4.02 0.063 0.134 8
5 { (period) p(t)} 417.302 5.68 0.027 0.058 9
6 { (a2) p()} 417.414 5.79 0.026 0.055 3
7 { () p(t + period)} 417.825 6.21 0.021 0.045 9
8 { () p()} 419.019 7.40 0.012 0.025 2
9 { (period) p()} 420.802 9.18 0.005 0.010 3
10 { () p(period)} 420.929 9.31 0.005 0.010 3
11 { () p(t period)} 422.266 10.64 0.002 0.005 11
12 { (t) p(t)} 423.331 11.71 0.001 0.003 14
13 { (t) p()} 423.561 11.94 0.001 0.003 8
14 { (period) p(t period)} 423.996 12.38 0.001 0.002 12
15 { (t) p(period)} 425.732 14.11 0 0.001 9
16 { () p(cohort t)} 439.523 27.90 0 0 29
Model (model 1). In general, those models in which survival was time-dependent
(models 68 and 1012) and which allowed capture probability to vary between
secondary periods (model 12) were not parsimonious or poorly t the data. Thus,
the best RD model had random and constant emigration probabilities, constant
survival and time-dependent capture probability between and within primary periods
(model 1). When comparing nested models, the LRT corroborated a constant survival
(model 4, P = 0.205) and random emigration pattern (model 2, P = 0.165), but it
suggested a temporal effect on the probability of an individual being unavailable for
capture (; model 3, P = 0.012).
Survival and Capture Probability
Apparent survival rates were high and stable during the study. The model-averaged
estimates of both approaches provided similar results: CJS: = 0.88 0.07 SE,
95% CI = 0.670.96; RD: = 0.89 0.03 SE, 95% CI = 0.820.94.
CJS models detected uctuations in recapture probabilities between cohorts
throughout the monitoring (Fig. 3). Capturability was lower for the 2004 cohort
(p =0.215 0.143 SE) and reached maximum values in 2003
(p =0.629 0.124)
and 2009 (p = 0.633 0.186).
Emigration Probability
We used model-averaged results from the RD models to investigate individual
movements. Specically, we estimated the probability that an individual available
= 0);
Markovian emigration (
(x)
(x) =
() =
() p(st) =
c(st)}
1,717.222 0 0.540 1.000 102
2 {()
()
(t) =
(t) p(st) =
c(st)}
1,719.735 2.510 0.154 0.285 108
4 {(t)
() =
() p(st) =
c(st)}
1,727.640 10.420 0.003 0.006 108
5 {()
(t)
(t) p(st) =
c(st)}
1,728.688 11.470 0.002 0.003 114
6 {(t)
()
() p(st) =
c(st)}
1,729.797 12.580 0.001 0.002 109
7 {(t)
(t) =
(t) p(st) =
c(st)}
1,731.990 14.770 0.000 0.001 114
8 {(t)
(t)
(t) p(st) =
c(st)}
1,744.131 26.910 0 0 120
9 {()
= 0 p(st) =
c(st)}
1,764.346 47.120 0 0 101
10 {(t)
= 0 p(st) =
c(st)}
1,772.300 55.080 0 0 107
12 {(t)
= 0 p(s) =
c(s)}
1,880.055 162.830 0 0 23
for capture on previous occasions temporarily emigrated from the study area (
),
and the probability that an individual that was outside the study area on a previous
occasion remained outside it (
= 0.33 0.07
SE; 95% CI =0.200.49). Thus, the probability of dolphins remaining in the study
area between capture occasions (1
=1
) to include
the unmarked individuals.
Proportion of
marked
dolphins RD abundance estimates
Year SE ()
N CV 95%CI
N
CV 95%CI
2002 0.53 0.065 37 0.027 3539 70 0.123 5589
2003 0.39 0.055 31 0.031 2933 79 0.143 60105
2004 0.24 0.060 25 0.251 1337 105 0.354 53204
2005 0.54 0.059 37 0.061 3241 67 0.117 5586
2006 0.59 0.091 34 0.163 2344 57 0.215 3887
2007 0.55 0.062 55 0.128 4169 100 0.169 72139
2008 0.38 0.079 34 0.144 2543 89 0.246 56144
2009 0.56 0.078 69 0.153 4890 124 0.201 83182
Trends in Abundance
The regression of the corrected abundance estimates (
N